Starting from some uncontrollable cases of current artificial intelligence (AI), the fragility and uncertainty of artificial intelligence are associated, and the methods of improving the controllability of artificial intelligence are tentatively proposed. The article is mainly divided into five parts. The first part briefly describes the proposal and development of AI. The second part introduces the specific uncontrollable cases of chat robots and other AI products. The third part mainly discusses the sources of AI technology fragility and uncertainty. The fourth part puts forward some suggestions for improving the controllability of AI, including enhancing the scenario awareness ability of AI, deploying controllable nodes in the system architecture and improving the supervision and evaluation system through the controllable nodes. The fifth part describes the prospects of AI technology in the field of optoelectronics.