Mapping Accountability in Human-AI Partnerships in Healthcare: Towards a Patient-centric Approach

Regulation and Accountability in the context of Human-Artificial Intelligence (AI) partnerships in Healthcare remain to be extensively studied, even as we witness the latter pushing the boundaries of assistive and augmented technology. In this context, we attempt to operationalise accountability in terms of its functional characteristics to assist in finding a fair manner of allocating the answerability among the Human-AI partners in health. Through the proposed Human-AI Accountability Assignment matrix (H3A Matrix), we show that the risk to patient safety is a dynamic attribute and the need to anchor on the micro-tasks as a concrete functional component to assign accountability. This division into micro-tasks allows us to unpack the ways in which different humans (including the patient-in-the-loop) interact with the devices. We argue that accountability evolves based on the contexts, stakeholder interaction and the scope of applications of AI-enabled devices and emphasise the need for an independent forum that encapsulates a mix of expertise and representation.