Artificial intelligence (AI) education for K-12 students is an emerging necessity, owing to the rapid advancement and deployment of AI technologies. It is essential to take teachers' perspectives into account when creating ecologically valid AI education programmes for K-12 settings. However, very few studies investigated teacher perception of AI education. Phenomenography is an empirical research method that was widely used to understand teacher's interpretive understanding of new phenomenon, in this study, the teaching of AI in secondary school. Therefore, the present study investigated teachers' conceptions of teaching AI using a phenomenographic approach. Twenty-eight in-service teachers from 17 secondary schools in Hong Kong were invited to participate in an interview after implementing an AI curriculum. Six categories of teacher conceptions were identified: (1) technology bridging, (2) knowledge delivery, (3) interest stimulation, (4) ethics establishment, (5) capability cultivation, and (6) intellectual development. The hierarchical relationships of the six concepts were organised as an outcome space. The space shows a range of surface to deep conceptions and offers an understanding of how teachers perceive AI education through their teaching experience. Two learning paths have been suggested for cultivating technical and non-technical teachers for teaching AI. These learning paths provide insights for teacher educators and policymakers to enhance teachers' competence in teaching AI and promote general AI education for K-12 students. Copyright © 2022 Springer.