With advances in devices and other technologies, the demand for immersive experiences that seamlessly blend the digital and physical worlds will be significant by the end of the decade. To support these, 6G networks will need to deliver unprecedented capacity, low latency, energy efficiency, and cognitive capabilities to manage vast radio resources. The potential of Artificial Intelligence (AI) to enhance the physical layer and reach these goals has been demonstrated in previous works, and now this paper explores the frontiers of Machine Learning (ML) on the wireless Medium Access Control (MAC) layer. ML techniques can further enhance the native Artificial Intelligence Air Interface (AI-AI) envisioned for 6G. This roadmap paper navigates recent research on AI-driven MAC functions such as resource allocation, random access, Adaptive Modulation and Coding (AMC), power control, protocol learning, Channel State Information (CSI) reporting, Hybrid Automatic Repeat Request (HARQ), and Multi-RAT Spectrum Sharing (MRSS).