Human centred explainable AI decision-making in healthcare

• HCXAI sets humans at the core of AI, understanding the socially situated nature. • The ethical risks of applying HCXAI in the healthcare domain are unknown. • It is unclear which explainability needs closely mimic human decision-making. • This research proposes an applied ethical HCXAI framework in healthcare. • The framework empowers professionals to make decisions in collaboration with AI. Human-centred AI (HCAI 1 1 HCAI – Human-centred artificial intelligence ) implies building AI systems in a manner that comprehends human aims, needs, and expectations by assisting, interacting, and collaborating with humans. Further focusing on explainable AI (XAI 2 2 XAI – Explainable artificial intelligence ) allows to gather insight in the data, reasoning, and decisions made by the AI systems facilitating human understanding, trust, and contributing to identifying issues like errors and bias. While current XAI approaches mainly have a technical focus, to be able to understand the context and human dynamics, a transdisciplinary perspective and a socio-technical approach is necessary. This fact is critical in the healthcare domain as various risks could imply serious consequences on both the safety of human life and medical devices. A reflective ethical and socio-technical perspective, where technical advancements and human factors co-evolve, is called human-centred explainable AI (HCXAI 3 3 HCXAI – Human-centred explainable artificial intelligence ). This perspective sets humans at the centre of AI design with a holistic understanding of values, interpersonal dynamics, and the socially situated nature of AI systems. In the healthcare domain, to the best of our knowledge, limited knowledge exists on applying HCXAI, the ethical risks are unknown, and it is unclear which explainability elements are needed in decision-making to closely mimic human decision-making. Moreover, different stakeholders have different explanation needs, thus HCXAI could be a solution to focus on humane ethical decision-making instead of pure technical choices. To tackle this knowledge gap, this article aims to design an actionable HCXAI ethical framework adopting a transdisciplinary approach that merges academic and practitioner knowledge and expertise from the AI, XAI, HCXAI, design science, and healthcare domains. To demonstrate the applicability of the proposed actionable framework in real scenarios and settings while reflecting on human decision-making, two use cases are considered. The first one is on AI-based interpretation of MRI scans and the second one on the application of smart flooring.