Press
Amsterdam, 23 July 2025. In this article published in Dutch health-tech platform SmartHealth, Rewire Partner Philipp Diesinger explains why bias detection is a prerequisite for safe and equitable AI in healthcare. When training data is unrepresentative, AI models systematically underperform for the groups least reflected in that data — women, older patients, and those from migrant backgrounds. Bias can enter at multiple points: data collection, measurement instruments, algorithmic proxies, and clinical data entry. Left undetected, these distortions lead to misdiagnosis, unequal treatment, and eroding trust. Diesinger argues that dataset audits, subgroup performance analysis, and fairness metrics must become standard practice in AI development, with human oversight maintained throughout.
