Artificial Intelligence in Education (AIEd) is an emerging interdisciplinary field that applies artificial intelligence technologies to transform instructional design and student learning. However, most research has investigated AIEd from the technological perspective, which cannot achieve a deep understand of the complex roles of AI in instructional and learning processes and its relationship with other educational elements. To fill this gap, this review research proposes a conceptual framework from complex adaptive systems theory perspective, uses a systematic literature review approach to locate and summarize articles, and categorizes the roles of AI in the educational system. The review results indicate that when AI is added into an educational system, its roles can be characterized into three categories: AI as a new subject, AI as direct mediator, and AI as a supplementary assistant to influence the instructor-student, student-self, and student-student relationships. Reviewed articles under each category are examined to understand the influences of AI applications on instruction and learning. Based on the results, this systematic review research proposes practical, theoretical, and technological implications of AIEd development.