Explainability in Human-AI Collaboration under Information Asymmetry – The Case of Surgical Duration Planning

As the integration of artificial intelligence (AI) into medical decision-making continues to expand, understanding its impact on decision-making performance is crucial. In the healthcare domain, inaccurate surgical duration forecasting poses challenges to optimal resource utilization. Although AI decision support capabilities are continually growing, humans typically remain responsible for final planning decisions – not only due to accountability and ethical concerns, but also because of information asymmetries, as physicians often possess contextual knowledge unavailable to the AI. Our study investigates how the integration of Explainable AI (XAI) affects acceptance and trust in AI-assisted decisions under symmetric and asymmetric information conditions. In a 2×2 experiment with healthcare professionals, participants received either AI predictions only or AI predictions accompanied by explanations (i.e., XAI) and each condition was tested in environments where either both the AI and the human had access to the same information (symmetric) or where the AI has access to less information than the human decision-maker (asymmetric). Results show that trust is lower in asymmetric environments and that XAI significantly improves trust in both environments. However, trust improvements through XAI do not consistently translate into forecasting performance gains. Our findings highlight the importance of understanding how information asymmetries shape trust in AI systems, and that XAI can positively influence trust under asymmetric conditions. At the same time, they caution against unintended effects of misaligned information structures in human–AI collaboration.