Advancing Trustworthy AI in Healthcare Through Meta-Research: Results of an Interdisciplinary Design-Thinking Workshop

Meta-research can bridge the gap between AI ethics principles and real-world healthcare practice. The paper identifies that while Trustworthy AI and meta-research share goals around evidence, robustness, and transparency, these fields rarely collaborate—despite meta-research offering concrete solutions to translate ethical principles into actionable healthcare applications.

Lack of common terminology is a major barrier to trustworthy AI in healthcare. The workshop identified "common terminology and understanding of TAI" as a key challenge, suggesting that healthcare professionals, AI developers, and ethicists may be talking past each other due to inconsistent definitions and language around trustworthy AI concepts.

Current evaluation metrics for AI in healthcare are inadequate. The paper highlights "choosing adequate evaluation metrics" as a pressing challenge, indicating that existing methods for assessing whether healthcare AI systems are trustworthy and performing well may not capture what matters most in clinical settings.

Reproducibility and transparency remain critical unsolved problems. The workshop identified ensuring "robustness, replicability, reproducibility" and addressing "lack of transparency" as major obstacles, suggesting that many healthcare AI systems cannot be reliably verified or understood by clinicians and regulators.

Real-world clinical validation is disconnected from AI development. The paper notes a gap between "preclinical biomedical research" and "validation in real-world clinical environments," indicating that AI systems developed in controlled settings often fail to work reliably when deployed in actual hospitals and clinical practice.