How and why science education scholars and practitioners might use artificial intelligence (AI) in the classroom has been a controversial agenda for decades. ChatGPT, a state-of-the-art (SOTA) AI released in November 2022, has attracted global interest for its exceptionally high performance in generating human-like natural language answers to almost any questions a user queries. Its ability to generate knowledge raises questions about what human knowledge means in science education in the digital age. The researchers of this study consider ChatGPT as an epistemic agent that is capable of generating new scientific knowledge and as an eligible interviewee to provide data for the study of AI and human episteme. The researchers interviewed ChatGPT and analysed its responses to questions about the future of human epistemic agency in science education to elicit ontological and relational implications between humans and AI. Based on ChatGPT's articulations, it was suggested that SOTA AI-generated knowledge is "non-bodied," "robust," and "detached," which can be summarised as "automatous." In contrast, it was suggested that human-generated knowledge is "embodied," "value-laden," and "engaged," which can be summarised as "personal," in light of Michael Polanyi's thoughts. As a result, this study insists that there are still unique characteristics of human-generated scientific knowledge compared to those of AI. Therefore, it is suggested that AI in science education should be understood as a collaborator in knowledge construction while securing the lead of human students in this process.