Last April, the Ministerial Delegation for Digital Health (DNS) published its recommendations for best practicesfor integrating ethics into the development of artificial intelligence solutions in health: implementing “ethics by design”. Here is a look at this report.
Brigitte Seroussi, head of the Digital Health Ethics Unit at DNS and university professor and hospital practitioner, heads the Digital Health Ethics Unit, which is organized into several working groups. Working group n°3, led by David Gruson, specialist and director of the Luminess health program, is at the origin of this methodological guide. Its goal is to integrate ethics from the development of AI solutions, which is called ” ethics by design “.
This report is the result of a consultation process with the various stakeholders in the digital health ecosystem. It is the culmination of a two-year process during which the work of the ethics unit involved multi-professional experts, including AI researchers, patient representatives, healthcare professionals, lawyers, sociologists, ethicists, etc. They include Emeline Brule (School of Engineering and Computer Science, University of Sussex), Madeleine Cavet (Radiologist), Caroline Guillot (Health Data Hub) but also Daniela Parrot (Data Protection Officer, Ministry of Solidarity and Health) and Philippe Lamoureux (LEEM). Together, they have established these recommendations of good practices, which aim to integrate ethics from the first stages of the development of AI solutions in health.
This white paper is addressed ” in priority “ to designers of solutions based on algorithms trained from massive data extraction, excluding self-learning evolutionary systems.
The development process of an AI solution includes many steps: scoping, data collection, data pre-processing, algorithm construction and algorithm evaluation before the digital solution is put into production. Ethical issues specific to each of these steps were proposed, which the experts referred to as “by design”. For example, suppliers will have to make a declaration of conformity and ensure that the high-risk AI system is CE marked before being put on the market. Each time it is modified, a new conformity assessment will have to be performed.
Some recommendations have already been incorporated into national and European law. We can quote in particular the law n° 2021-1017 of August 2, 2021 relating to bioethics, which introduced an obligation of information for the professionals using an AI in health matters.
Translated from La DNS publie ses recommandations pour une éthique de l’IA « by design » pour la santé
