McGill U. - Desautels Faculty of Management, Canada
Artificial intelligence (AI) technologies differ from previous organizational technologies in their ability to autonomously engage in complex knowledge work. Crucial to this capability is the incorporation of specialized knowledge during development. However, developing early instances AI for complex cross-boundary work presents challenges in determining essential knowledge and how it should manifest in the technology's features and usage behaviors. To explore this, I conducted a two-year qualitative field study on the concurrent design and development of AI-based scheduling support tools for Operating Rooms (ORs) at two hospitals. A single technology vendor collaborated with teams from each hospital, adapting the same early prototype to local conditions. Based on 96 hours of observation and 77 interviews, I identified how local differences in knowledge perceptions and arrangements can drive divergences in the technology's development path. This study contributes to a processual ontology of artificial intelligence by showing that failing to recognize and include all clinical and processual knowings associated with OR scheduling results in an assembly of somewhat successful technological elements that do not ultimately cohere into an effective AI system capable of participating in complex cross-boundary healthcare processes.
