An AI-based roof fall hazard detection system is developed for the future of mine autonomy. Subtropolis Underground Limestone Mine (SULM) is used as a case study. SULM experiences routine roof falls caused by high horizontal stresses. High horizontal stresses produce visual clues in the form of roof beams since the roof beams subjected to horizontal stresses deform or sag more than those only subjected to gravity. The ground control personnel in the SULM have been able to associate areas of high horizontal stress with roof beams. They have identified the depth and the frequency as the most important characteristics of the stress-induced roof beams. The association between the characteristics of roof beams with roof fall hazards is also verified by a statistical analysis of past roof falls and the frequency–depth measurement of roof beams. Since the roof beams are visual clues, the researchers decided to work with images, and chose convolutional neural networks (CNN) as the AI approach. Therefore, based on the recommendations of ground control personnel, images depicting hazardous and non-hazardous roof conditions are collected.