Constitutive laws describing frictional response at rock interfaces have challenged modelers for decades. Since the pioneering work of Dieterich, Ruina, and Rice in the 1980s, a variety of rate-and-state friction laws have been proposed. Despite their successes, significant difficulties remain in formulating models that accurately capture interface behavior across wide ranges of sliding velocities. In this talk, we review recent efforts to leverage machine learning (ML) techniques to efficiently “learn” friction laws, with an emphasis on both numerical efficiency and physical fidelity. We discuss approaches that use data-driven learning to infer frictional behavior [4], as well as the development of neural surrogate models that can be embedded within variational numerical frameworks. We will present prior work employing neural networks and ongoing research based on neural operators, highlighting their potential to generalize across regimes and numerical discretization.