Unmanned Aerial Vehicles (UAVs) have been used for live streaming for broadcasting large events, conducting real-time surveillance, search and rescue due to their high mobility and outstanding unmanned operation ability. However, UAV-based live streaming presents challenges due to frequent network fluctuations caused by the mobility of UAVs and diverse Quality of Service (QoS) and Quality of Experience (QoE) requirements. Although multipath transmission technology is an important enabler of the transport layer, its traditional rule-based multipath transmission mechanisms suffer from performance degradation when dynamic network conditions are present. Artificial Intelligence (AI) technology is capable of assisting multipath transmission mechanisms with more adaptive and flexible decision-making when dealing with network fluctuations and diverse QoS and QoE requirements. Furthermore, we propose Smart-Flycast, the first AI-driven multipath transmission UAV-based live streaming system. Specially, to improve QoS and QoE in the highly mobile UAV flight environment, we design three novel learning-based multipath transmission mechanisms. Our proposed solution may inspire other possible research directions utilizing advanced AI and multipath transmission technologies.