Title: Towards Safer Medical AI Reasoning and Action

Abstract: Multimodal generative AI (GAI), particularly multimodal LLMs, is rapidly transforming the landscape of clinical AI. Yet translating this promise into real-world healthcare remains challenging, as model reliability and safety are tested by the complexity of clinical workflows, heterogeneous data, and diverse medical practices. In this talk, I will share our research on key limitations of current GAI systems, including skewed reasoning over multimodal clinical information, declining performance as clinical action spaces grow, and biases arising from regional diXerences in medical practice. I will also discuss our approaches to strengthening the safety and reliability of these AI systems in clinical decision-making and healthcare delivery.

Biography: an Assistant Professor of Personalized Medicine at MBZUAI. He received his Ph.D. from Imperial College London, UK, in 2023. His current research interests include medical foundation models, agents, and human-AI interactions. Some of his work has been published in leading journals such as NEJM AI, Nature Machine Intelligence, and The Lancet Digital Health. He serves as a guest editor for IEEE Journal of Biomedical and Health Informatics.