The 5th wireless communication (5G) techniques not only fulfil the requirement of 1, 000 times increase of internet traffic in the next decade, but also offer the underlying technologies to the entire industry and ecology for internet of everything. Compared to the existing mobile communication techniques, 5G techniques are more-widely applicable and the corresponding system design is more complicated. The resurgence of artificial intelligence (AI) techniques offers as an alternative option, which is possibly superior over traditional ideas and performance. Typical and potential research directions to which AI can make promising contributions need to be identified, evaluated, and investigated. To this end, this overview paper first combs through several promising research directions of AI for 5G, based on the understanding of the 5G key techniques. Also, the paper devotes itself in providing design paradigms including 5G network optimization, optimal resource allocation, 5G physical layer unified acceleration, end-to-end physical layer joint optimization, and so on.The 5G technology standards are in development, getting complete and mature [2,3]. In December 2017, 3G Partnership Project (3GPP) officially announced the new standards for 5G New Radio (NR) which include supports for 5G Non-Standalone architecture (NSA) and eMBB [4]. On June 14, 2018, 3GPP formally completed the Standalone (SA) version of the 5G NR standard, marking a long-awaited target date for 5G standardization [5]. These announced standards effectively set the stage to launch full-scale and cost-effective development of 5G networks. Compared to the current 4G networks, 5G NR: (1) enhances the MIMO systems with the massive MIMO technology; (2) makes completion to the time slot structure and resource block (RB) allocation of the orthogonal frequency-division multiplexing (OFDM), proposing a more flexible air interface; (3) will introduce the non-orthogonal multiple access (NOMA) to support the Internet of Things (IoT) in the near future; (4) follows the previous distributed antenna systems [6], splits the wireless functions into distributed units (DU) and central units (CU) and applies network virtualization and network slicing techniques based on cloud computing.Overall, 5G networks will tailor the provisioning mechanisms for more applications and services, hence is more challenging with the complicated configuration issues and evolving service requirements. Before 5G, researches of communication systems mainly aim at satisfactory data transmission rate and supportive mobility management. In the 5G era, the communication systems will gain the abilities to interact with the environment, and the targets are expanded to joint optimizations of ever-increasing numbers of key performance indicators (KPIs) including latency, reliability, connection density, user experience, etc [7]. Meanwhile, new features like the dynamic air interface, virtualised network and network slicing introduce complicated system design and optimization requirements as long as ...