The Pulse Pioneering AI: Professor John Kelleher, his vision and work In February 2024, John Kelleher became Professor of Computer Science at Trinity’s School of Computer Science and Statistics and the Director of the SFI-funded ADAPT Research Centre (www.adaptcentre.ie) John, you did your PhD in AI over two decades ago, even before it was fashionable. What motivated you then to do your doctorate in AI? Although my formal training was in computer science, I have always been deeply interested in philosophy and cognitive science. Pursuing a doctorate in AI allowed me to bring these three areas together. Questions about perception, attention, memory, and the sources and representation of knowledge have been central themes throughout my research career. My early work focused on the interface between language and perception – specifically, how knowledge is represented at this interface and how it is mediated by attention and memory. These questions laid the foundation for my broader exploration of language models and machine learning. Even today, the core challenges of how a computational model represents information and distributes attention across it when making decisions remain at the heart of AI research. These issues are especially critical in safety-sensitive scenarios, where understanding the factors that lead an AI system to generate a particular output is vital. This continuity in the questions I’ve pursued – from the early days of AI to the modern era – reflects how foundational and enduring these challenges are in the field. What do you think is the most transformative development in AI/generative AI in the health/ health system space in recent years? The field of AI in healthcare is evolving rapidly across multiple fronts, making it difficult to pinpoint just one transformative development. However, several key advancements stand out to me. First, integrating AI systems into clinical workflows has been revolutionary, particularly in supporting clinical documentation and enabling personalised patient interaction. For instance, these systems are now being used to capture patient-reported outcome measures, offering insights that were previously hard 4 Professor John Kelleher, Director of the ADAPT Research Centre to gather at scale. This is largely made possible by the sophistication of large language models (LLMs), which can process and generate language in a way that feels natural and intuitive for human (non-expert) users. In the realm of data, AI is playing a critical role in synthesising datasets to enable training and research while preserving patient privacy. This is especially valuable for advancing medical research without compromising confidentiality. Personally, I find the ability of modern AI systems to process multi-modal data (combining text, images, and other formats) and longitudinal data (tracking changes over time) particularly transformative. These capabilities have immense potential for creating advanced clinical decision support systems, which can significantly enhance personalised medicine by tailoring treatments to individual patients. However, as promising as these developments are, realising their full potential requires addressing substantial challenges. Ethical considerations, Winter Edition 2024
Download PDF file