Scientific machine learning combines data-driven methods with domain knowledge, simulation and computing resources. Its principles offer a useful framework for evaluating AI in complex technical operations, from data quality and model reliability to automation and human decision-making.