Scientists have developed a novel machine-learning framework that can automatically map out phase diagrams for novel physical systems without requiring large, labeled training datasets. This physics-informed approach leverages generative artificial intelligence models to estimate the probability distribution of measurement statistics in a physical system, creating a classifier that can determine the phase of a system given certain parameters. This method is computationally efficient, works automatically without extensive training, and can help scientists discover unknown phases of matter autonomously, with significant implications for materials science, quantum computing, and thermodynamics.