How AI Helps to Compile Human Intelligence: An Empirical Study of Emerging Augmented Intelligence for Medical Image Scanning

Abstract

Artificial intelligence (AI) is advancing continuously. However, full delegation to an AI application is often not possible or desirable due to technical limitations, ethical concerns or legal issues. Augmented intelligence systems, where humans and AI work together jointly, have been proposed to improve decision making in complex, uncertain and failure‐intolerant environments. Yet, this raises questions about how compatible human and AI knowledge are, and whether translating between the two increases decision making intelligence, or whether it effectively limits AI applications' capacity for computational agency and human agents' capacity to consider uniquely human knowledge. We explore this notion by looking at augmented intelligence in terms of systemic intelligence and mutual learning. Building on an emergence perspective, we perform a case study of an augmented intelligence system for image‐based diagnostics in the radiology branch of a medical care centre. Our findings indicate a strong distinction between specialists' and non‐specialists' intelligence augmentation with AI. This distinction fuels generative cycles which produce iteratively more sophisticated algorithms, human representations and practical routines. Drawing on this analysis, we propose three stages by which new forms of intelligence emerge from the addition of AI recommendation tools, specifically, intelligence by propagation, intelligence by specialisation and intelligence by articulation.