The proposed research will focus on analyzing the current legal framework of AI’s transparency in healthcare and finding the technically feasible way to achieve it. While the opacity challenge of AI is related to the ‘black-box’ nature of AI algorithms, the solution shall be developed in cooperation between data scientists and legal researchers. For that, the proposed project will involve scientific centers from different areas: Law, Science, Technology and Society Research Group (LSTS) from Vrije Universiteit Brussel and the ETIS Research Laboratory (ETIS Lab) from CY Cergy Paris University. The models to explain AI (both ex-ante and ex-post) developed and tested in the ETIS lab will be explored and described in the context of transparency as a legal concept.
Further, the transparency issue and identified measures to achieve it (both compliant from legal and technical points of view) will be explored in the healthcare context. The research is based on the hypothesis that the transparency of AI in healthcare is important but not an absolute requirement and shall be always balanced with its safety and quality (accuracy). According to data scientists, the most advanced algorithms are often the most accurate and at the same time least explainable. Thus, the safety and performance of AI might be the trade-offs of its full transparency. Additionally, healthcare
itself is the domain where the highest level of transparency is hardly achievable. Any treatment is always a complex, risky and unpredictable process. While transparency is a crucial element for building trust in the use of AI in healthcare, requiring full transparency would put too extensive burden on AI that might diminish its benefits. To find the right level of AI’s transparency balanced with its safety and quality, the relevant legal sources and scholarly papers will be described, classified, compared and evaluated. Based on that, the solution balancing AI’s transparency with its safety and performance will be suggested and recommendations to improve the law will be provided.
Effective start/end date1/10/2030/09/24

    Flemish discipline codes

  • Artificial intelligence not elsewhere classified

    Research areas

  • artificial intelligence, transparency, healthcare, explainability, safety, quality

ID: 53167761