How climatic changes affect atmospheric blocking events, which are associated with extreme weather, is difficult to assess due to climate model uncertainties and the phenomenon's inherent complexity. To address this, I will discuss Deep Learning models that infer blocking frequencies using paleoclimate records. I will show that the models, despite not using direct inputs from paleoclimate proxies or knowing their exact locations, effectively maps the impact of historical climate shifts onto changes in blocking patterns. This study reveals that significant variations in blocking activity align with shifts in tropical Pacific conditions and El Niño Southern Oscillation (ENSO) events. This research underlines the need for accurate modeling of tropical Pacific climate to improve present-day simulations and future projections of atmospheric blocking. By using deep learning to interpret paleoclimate data, we can better understand how external forcings have historically shaped weather patterns and refine future climate projections.