Smart Experiments: An Introduction to How AI Is Transforming Modern Scientific Discovery

We are creating a new era of discovery by developing a unified suite of machine-learning frameworks that fundamentally transform how large-scale experimental facilities operate and extract scientific insight. Our work introduces an AI “beamline scientist” that can help design and steer complex neutron and x-ray experiments in real time; adaptive neural networks that learn directly from experimental data how to separate true physical signals from complex or structured backgrounds; and innovative implicit models—including wavelet-based architectures—that compress rich scientific data and reveal hidden parameters that govern material behavior, or that can learn from measured spectra how to extract parameters from the 'so-called' Hamiltonian used to model a given system. We furthrmore show how these implicit neural networks can be used to tackle major imaging challenges, such as reconstructing single biomolecules directly from raw data. Furthermore, we will introduce work that uses generative AI with equivariant networks to predict new crystal structures, develop fast surrogate simulators to guide experiments on the fly, and design machine-learning methods that dramatically speed up x-ray detector processing. Together, these advances show how AI can accelerate scientific progress, expand what experiments can measure, and open entirely new frontiers in understanding the natural world.