Abstract

Artificial intelligence offers a new path to chemical processes: enabling chemical systems that not only learn from data but also improve themselves over time. In this talk, I will outline how AI can close critical gaps in process development and operation, paving the way towards autonomous, self-optimising laboratories and process plants. I will focus on three complementary algorithmic approaches. The first, Bayesian optimisation, provides a framework for experimental design and optimisation. I will particularly discuss how we can use multi-fidelity and human-in-the-loop strategies for a more informed process optimisation. Second, I will discuss how large language models (LLMs) can and are being leveraged to capture human knowledge, interact with algorithms, and enable autonomous processes via symbolic reasoning and algorithmic search. Finally, I will talk about how reinforcement learning, famous for making computers “learn” and beat the best humans in the world at various tasks, unlocks new opportunities for control and real-time optimisation of complex, dynamic processes, with particular promise in bioprocess applications. Together, these advances point to a unifying vision: AI-driven chemical processes that are faster to develop, safer to operate, and inherently more sustainable. I will conclude by reflecting on the emerging paradigm of autonomous process innovation and the opportunities it creates for the chemical sciences and industry.