When AI Regulation Meets Reality
Legal and ethical frameworks for AI are often created at a distance from the environments where AI systems are actually developed. While these regulations aim to ensure fairness, transparency, and safety, developers frequently experience a gap between regulatory expectations and the practical realities of building AI models.
As part of WP1, discussions with AI development teams explore how legal and ethical requirements are experienced during the development process, particularly in healthcare and data-driven research.
One key theme is that developers generally understand the importance of regulation, especially regarding patient protection, fairness, and accountability, and requirements around explainability, documentation, and bias mitigation are to a large degree perceived as well aligned with scientific quality expectations.
Data-related requirements are among the biggest concerns. Ensuring data quality, completeness, representativeness, and security is critical for trustworthy AI, but real-world healthcare data is often incomplete, fragmented, or potentially biased. Developers therefore rely on mitigation strategies such as transparency about limitations, human oversight, and continuous monitoring.
Legal requirements also increasingly shape model design itself. Expectations regarding automatic logging, transparency, and human oversight are generally viewed as valuable safeguards, but they can also introduce technical complexity and slow development processes.
Questions about responsibility, acceptable risk, and fairness remain difficult to answer in practice. The discussions further highlight the importance of involving clinicians, patients, and other stakeholders in AI development. Including those who will ultimately use or be affected by AI systems can improve both ethical robustness and practical usability.
Overall, the conversations within WP1 demonstrate that ethical and legal frameworks are becoming an integral part of AI development. The challenge moving forward is to ensure that these frameworks not only protect individuals and society, but also remain realistic, understandable, and adaptable to the realities of innovation.