Cognitive science studies the mind through behavior; neuroscience asks what mechanisms inside the brain give rise to that behavior. A century of probing, recording, and lesioning has revealed that complex cognition arises from structured neural systems — specialized regions, modular networks, and characteristic circuits — rather than from a homogeneous substrate. Today’s neural language models pose an analogous question: behind their behavior lies a vast network of weights, activations, and attention patterns, but what kind of computational structure actually gives rise to what we observe? Mechanistic science of AI seeks to open this black box, drawing methods from systems neuroscience and interpretability to identify the circuits, representations, and algorithms inside models that produce intelligent behavior.