Watch the event recording.

Many of the most important systems in medicine and science, from cellular differentiation and disease progression to modern AI models, can be viewed as complex, partially observable dynamical systems. Their internal mechanisms are hidden, yet they drive high-stakes decisions.

In this talk, Dr Chris Banerji will present a mathematical framework for 'explaining the unseen', drawing on information theory and differential geometry to characterise constraints on how such systems evolve. He will show how quantities such as entropy, curvature, and manifold flow reveal structure in biological processes and in data as it propagates through black-box AI models.

He will then turn to clinical AI, focusing on interpretable, concept-based approaches for real decision-making in cancer care. While these models promise transparency, they introduce challenges such as information leakage and accuracy–interpretability trade-offs, which can again be analysed and mitigated mathematically. Dr Banerji will conclude by showing how interpretable AI can enable concept discovery and scientific insight, positioning it as both a clinical and scientific tool.

This event is part of the King’s Institute for Artificial Intelligence’s AI Frontiers series.

Meet the speaker

Dr Chris Banerji is an AI+ Senior Fellow (Clinical–Academic) and clinician–scientist working at the intersection of artificial intelligence, mathematics, and medicine. He studied Mathematics at the University of Oxford, completed a PhD in computational and cell biology at UCL, and trained in medicine at Imperial College London. He has held postdoctoral positions at King’s College London and the Alan Turing Institute alongside clinical training at St Thomas’ Hospital and University College London Hospitals.

His research focuses on human-centred, trustworthy AI, combining concept-based models, conformal prediction, and human-in-the-loop decision support for high-stakes clinical workflows. His primary applied interests are histopathology and oncology, developing AI to support cancer diagnostics and multidisciplinary decision-making. He also maintains an active programme in neuromuscular disorders, particularly facioscapulohumeral muscular dystrophy. Methodologically, his work draws on dynamical systems, differential geometry, and information theory, with a parallel commitment to interdisciplinary education and AI governance.