Thomas Souverain

Postdoctoral Fellow @CEA

Paris, FR

I’m the responsible for ethics & conformity to the #European AI Act in the #OpenLLM consortium developing a frugal, open and sovereign European Large Language Model (LLM) in education. In the context of French education, I lead field studies on the impact of LLM on the role of high-school teachers in their classrooms. I’m also developing advanced coding tools to explain and grant safety of these LLM models. The consortium involves the prestigious research institutions in Computer Science of the French Alternative & Atomic Energy Commission hosting my works (CEA Paris Saclay), Polytechnique, LORIA and Panthéon Sorbonne, and pioneer actors of open models such as Linagora and Opsci.ai. Still as a philosopher, I keep on my research on AI fairness and explanation. For 5 years, I have been creating an original teaching on AI Ethics for Master’s students in Computer Science (CentraleSupélec, USPN), in Digital Law (ICP) and in Philosophy (ENS). I make them aware of ethical issues from privacy to biases, mixing in-depth philosophy readings, debates and play roles inspired by my personal experience of AI in loan lending, recruiting and education. Beside my students, I also share my knowledge and experience on AI ethics to companies, designers and deployers. As a European expert, I contribute to the writing of European standards of conformity to the AI Act. I have been piloting for 4 years the AI Ethics Group of the French association of AI practitioners (Hub France IA), where I coordinated two open white papers, with concrete guidelines and checkboxes of best practices to implement ethical perspectives in AI designs.