The Earth's climate system has undergone numerous abrupt transitions in the past, with drastic changes occurring over just a few decades. Such shifts arise from tipping points, which are critical thresholds beyond which the system reorganizes, often abruptly and sometimes irreversibly. Anthropogenic greenhouse gas emissions constitute a major perturbation to the climate system and have the potential to trigger future tipping points, including carbon-cycle feedbacks that may amplify the initial warming. Understanding tipping dynamics requires extensive exploration of a model's parameter and state space, which in turn demands large numbers of simulations. While this is feasible for conceptual models with few parameters and negligible computational cost, it is impractical for Earth system models, which are characterized by high dimensionality and substantial computational expense. This research bridges the gap between conceptual and complex models by using machine learning to reduce a complex Earth system model to a low-dimensional conceptual representation. The resulting system of ordinary differential equations is then analyzed to uncover the dynamical properties that give rise to tipping points under past and future climate conditions. My presentation will demonstrate the potential of this approach for identifying climate tipping dynamics in complex models and for advancing our understanding of long-term climate behaviour through the integration of Earth system modelling, machine learning, and dynamical systems theory.