Designing and Implementing AI in Healthcare: Experiences with Sociotechnical Approaches

AI-based tools may improve the precision and appropriateness of healthcare, ease the synthesis of complex information, and reduce the burden of clinical tasks. As with other clinical informatics interventions, the design and implementation of AI-based tools is likely to reconfigure a wide array of work practices, roles and responsibilities. Sociotechnical strategies for understanding clinical processes are vital to ensure that novel tools are responsive to the complex realities of healthcare. These complexities include factors like the tension between the team-based nature of clinical work and the individualized nature of computer use; the need to navigate diverse information sources; uncertainty about the validity of clinical information; and often significant time pressure. Sociotechnical approaches support the development of AI-based tools that are transparent and trustworthy, and that augment, rather than disrupt, clinical practice by drawing on theories and methods from cognitive psychology, human factors engineering, anthropology and sociology, organization studies and other social sciences. A significant portion of current AI tool development focuses on diagnostic or other “traditional” clinical decision support use cases, with the promise of improved accuracy and precision over rule-based tools. Supported in part by large language models, tools beyond traditional point-of-care decision support have recently exploded, as well. These include applications like conversational agents for patient education and navigation, ambient transcription and rapid phenotyping in genetic testing pathways. Despite the proliferation of use cases, the empirical literature describing sociotechnical approaches to the design and implementation of AI-based tools remains limited. The goal of this workshop is to share real-world experiences with the design and implementation of AI-based tools in clinical settings.