Soil erosion is a critical global challenge, driving land degradation, reducing agricultural productivity, and disturbing ecological balance. Water-induced erosion affects vast landscapes and is expected to worsen under climate change. Empirical models such as the Universal Soil Loss Equation (USLE), its revised form (RUSLE), and regional adaptations have long served as essential tools for assessing erosion risk and guiding conservation strategies. Their computational simplicity and applicability in data-scarce regions make them widely used, but reliance on empirically derived coefficients limits their ability to capture the complex, non-linear interactions of soil, vegetation, and atmosphere. Recent advances in machine learning (ML) are transforming erosion modelling by incorporating diverse datasets, including meteorological records, remote sensing indices, topography, and land use patterns. Unlike classical models, ML and deep learning approaches can model spatiotemporal dynamics and provide near-real-time erosion risk assessments with higher accuracy. Hybrid approaches that integrate empirical frameworks with AI-driven enhancements are emerging as powerful tools for sustainable land management. This chapter reviews the evolution, calibration, and limitations of empirical soil erosion models and highlights the paradigm shift introduced by ML. It concludes with future directions for integrating empirical knowledge and AI solutions to strengthen erosion prediction and support climate-resilient agriculture.