14/08/2026
How AI agents are reshaping the future of scientific discovery!
We were pleased to host Prof George Karniadakis for an engaging seminar on Scientific Machine Learning (SciML) and the growing potential of agentic AI systems in tackling complex challenges across science and engineering.
Prof Karniadakis presented AgenticSciML, a collaborative multi-agent framework in which more than 10 specialised AI agents interact to generate, evaluate, and improve scientific machine learning solutions. Through structured debate, shared memory of methods, and evolutionary search mechanisms, the system enables exploration of novel model architectures and optimisation strategies beyond traditional expert-driven design.
Across physics-informed and operator learning tasks, AgenticSciML demonstrated significant improvements over single-agent and human-designed approaches, while also discovering novel strategies such as adaptive mixture-of-expert architectures, decomposition-based physics-informed neural networks (PINNs), and physics-informed operator learning models.
The session offered a fascinating glimpse into how collaborative reasoning among AI agents could enable more scalable, transparent, and autonomous approaches to scientific computing and discovery.
Thank you to everyone who joined us for this thought-provoking session and engaging discussion!
