Abstract: With the deepening development of "artificial intelligence+", more and more work requires collaboration between humans and AI. However, previous studies have mostly explored the impact of AI systems on human-AI collaboration from a technical perspective, but have paid less attention to the fact that with the increasing autonomy of AI technology, the relationship between humans and AI is shifting from "tools" to "teammates". Faced with the challenges brought about by the transformation of this relationship model, how to ensure the effectiveness of human-AI collaboration has become a key issue that urgently needs to be addressed.
In the face of this issue, this paper breaks away from the previous practice of viewing AI and human interaction unilaterally from the perspective of AI technology systems, and instead incorporates humans into the analysis system, focusing on the four dimensions of human-AI collaborative autonomy: adaptive learning, uncertainty, opacity, and openness. Based on role theory, three studies were conducted using system screening method, content analysis method, critical synthesis method, etc. Study 1 points out that the uncertainty dimension and opacity dimension of human-AI collaborative autonomy lead to role conflicts that hinder human-AI collaboration, aiming to explore how the two dimensions of uncertainty and opacity can ultimately reduce human-AI collaboration through role conflicts and human-AI fit; Study 2, from the perspective of "AI actively adapting to human roles", points out that the openness dimension will enable AI systems to further understand and grasp the technical characteristics and role requirements required for collaboration with humans through role taking, thereby enhancing AI's resilience and enabling AI to better understand human intentions and collaborate; Study 3, from the perspective of "humans are willing to accept AI as a 'new teammate'", points out that the adaptive learning dimension will further enhance their anthropomorphism through role learning, allowing human teams to better understand and accept AI, ultimately achieving the goal of promoting human-AI collaboration. Three studies complement each other and construct a human-AI collaboration model, aiming at exploring how human-AI collaborative autonomy ensures effective collaboration between humans and AI.
Based on previous research, this paper has three main innovations: firstly, it extends the research boundary of AI technology (automation → autonomy) and introduces the concept of "human-AI collaborative autonomy". Compared with other technological features, the human-AI collaborative autonomy breaks away from the shortcomings of exploring the interaction between AI and humans unilaterally and through a static logic of mutual substitution. Instead, it takes the human-AI joint cognitive system as the basic analysis unit, including humans in the analysis system, emphasizing the collaborative complementarity, interactive trust, and dynamic balance between both parties. Secondly, it broadens the research perspective on the interaction between humans and AI (technical perspective → coupling perspective). Most existing research explores human-AI interaction from a technological perspective, focusing on the impact of the advantages of AI technology on human work and life. Based on the coupling perspective of existing research and role theory, we believe that the collaboration between humans and AI requires not only "AI actively adapting to human roles as an independent individual", but also "humans willing to accept AI as a new teammate. Both are indispensable. Thirdly, it enriches the relationship pattern between humans and AI (tool usage → teammate collaboration). Previous explorations on the relationship between humans and AI mostly placed AI in an auxiliary position, believing that AI is a tool that requires human operation. This paper breaks the previous recognition of the relationship between humans and AI, and argues that with the increasing advancement of technological autonomy, the relationship between humans and AI has gradually shifted from the tool usage to the teammate collaboration. AI has begun to integrate human teams as an independent intelligent entity, interdependent with human team members on an equal footing and playing an irreplaceable role in the team.

Key words: human-AI collaborative autonomy, human-AI collaboration, role conflict, role taking, role learning