Sitting at the intersection of wireless communication and ML, the talk will focus on two important aspects of wireless edge AI. First, we will discuss and demonstrate the application of ML in wireless communication for understanding, orchestrating, securing and maximizing the use of spectrum resources through learning. ML techniques can provide significant leaps in performance and efficiency of key L1 functions surrounding channel sensing, channel modeling, modulation and receiver design, and spatial re-use, as well as improving access and coordination schemes. We will explore how some of these ideas are advancing the 5G RAN today and how they can evolve to enable 6G.Second, we describe the role of Distributed Edge AI in the wireless environment. Owing to the distributed nature of data arising from sensors, base stations, and so forth, the goal in edge AI is to train privacy-preserving machine learning models under resource constraints. We provide an overview of recent techniques such as federated learning, distillation and split learning. We will also explore how to harness over-the-air computing and analog communication to provide scalable and privacy-preserving over-the-air model training. The talk will conclude by shedding light onto the next frontier of edge AI sitting at the confluence of semantic communication and ML.