AIMS: An uncertainty-aware AI experimentalist for quantum matter

Autonomous scientific agents are beginning to accelerate discovery, but most demonstrations operate in digital or highly structured settings where the objects, actions, and objectives are largely predefined. Quantum materials experiments pose a harder problem where uncertainties involve: the instrument state can drift, the useful signal may occupy only rare regions of an inhomogeneous sample, and the physical mechanism is often under-determined. Here we introduce the AI agent for Inference and Measurement in Science (AIMS), an uncertainty-aware closed-loop AI experimentalist for cryogenic microwave impedance microscopy that converts uncertainty into experimental action. AIMS links three nested loops: navigation under uncertain perception, measurement selection under sample inhomogeneity, and scale-resolved mechanism attribution under ambiguous physics. In navigation, it relocates the sample after cryogenic displacement, flags unreliable position estimates, and invokes recovery strategies, significantly reducing sample-locating time. In measurement, it maps twist angle distribution and generalized Wigner crystal score of twisted bilayer MoSe$_2$ to identify regions with the strongest correlated response. In discovery, AIMS asks not whether melting is simply classical or quantum, but how the competition of Coulomb repulsion, hopping, and other energy scales shapes the observed hierarchy. By testing the classical limit, varying hopping and Coulomb scales, and preserving sample morphology as a secondary testable variable, AIMS prioritizes a quantum-fluctuation-renormalized origin of the anomalously robust $\nu = 1/2$ crystal. AIMS demonstrates uncertainty-aware experimental agency for quantum matter with perception recovery, measurement choice, and energy-scale-resolved mechanism attribution in one closed loop.