Explainable Artificial Intelligence Methods in Combating Pandemics: A Systematic Review

Transparency is critical for real-world AI adoption in healthcare: The paper finds that lack of model transparency was a major reason why many AI solutions for COVID-19 failed to have clinical impact, despite numerous peer-reviewed demonstrations of their technical capabilities. This suggests that even well-functioning AI systems won't be used by healthcare providers unless they can understand how the system reaches its conclusions.

Explainable AI (XAI) can simultaneously improve performance and build trust: The review identifies that successful implementation of XAI techniques can achieve dual benefits—not only do they help AI models perform better, but they also increase end-user trust and influence actual decision-making by clinicians, addressing a key barrier to adoption.

XAI techniques require proper evaluation to deliver value: The paper emphasizes that evaluating XAI results is an important step that shouldn't be skipped, implying that simply adding explanations to AI systems isn't enough; those explanations must be validated to ensure they actually help users make better decisions.

Practical implementation guidance matters: The authors provide a checklist of suggestions for the experimental design process and address common implementation challenges with specific solutions, recognizing that the gap between developing AI and deploying it successfully requires attention to real-world factors beyond just the technology itself.