John A. Lee : From stochastic parrots to artificial general Intelligence (conférence en français)

Samedi 25 avril 2026 à 10h15 - Séance publique consacrée à l'impact des avancées technologiques en thérapie oncologique From stochastic parrots to artificial general Intelligence. The case of radiation oncology and proton therapy John A. LEE (FNRS - IREC - UCLouvain) Radiation oncology relies heavily on computerized technologies and is multi-disciplinary, with influences from physics, mathematics, medical imaging, and, more recently, machine learning and artificial intelligence. Its ultimate goal is to preserve the optimal balance between maximal effectiveness and minimal side effects, over the whole course of treatment, which can take several weeks, from initial planning to actual end of delivery. For that purpose, AI comes as a strong asset to automate steps in treatment planning and delivery, ensuring consistency and low variability, allowing for complex and adaptive protocols. For now, AI is capable of memorizing subtle nuances across patients in a population, retrieving and adapting those on demand when presented with a new patient. Although extremely powerful for information retrieval, the current generation of AI struggles at formal reasoning. Agentic AI and artificial general intelligence (AGI) are intended to break this glass ceiling, making the next generation of AI more than knowledgeable, capable of taking sequence of actions, and effectively turning it into a personal assistant. In radiation oncology and proton therapy, this means that treatment planning could be fully automated, with agents capable of carrying out autonomously most of the interaction with the treatment planning system and hardly requiring any control or supervision.