Artificial Intelligence (AI) is transforming healthcare and shows considerable promise for the delivery of medical education. This systematic review provides a comprehensive analysis of the global situation, effects, and challenges associated with applying AI at the different stages of medical education. This review followed the PRISMA guidelines, and retrieved studies published on Web of Science, PubMed, Scopus, and IEEE Xplore, from 1990 to 2022. After duplicates were removed (n = 1407) from the 6371 identified records, the full text of 179 records were screened. In total, 42 records were eligible. It revealed three teaching stages where AI can be applied in medical education (n = 39), including teaching implementation (n = 24), teaching evaluation (n = 10), and teaching feedback (n = 5). Many studies explored the effectiveness of AI adoption with questionnaire survey and control experiment. The challenges are performance improvement, effectiveness verification, AI training data sample and AI algorithms. AI provides real-time feedback and accurate evaluation, and can be used to monitor teaching quality. A possible reason why AI has not yet been applied widely to practical teaching may be the disciplinary gap between developers and end-user, it is necessary to strengthen the theoretical guidance of medical education that synchronizes with the rapid development of AI. Medical educators are expected to maintain a balance between AI and teacher-led teaching, and medical students need to think independently and critically. It is also highly demanded for research teams with a wide range of disciplines to ensure the applicability of AI in medical education.