关键词: ChatGPT；算法权力；主体性；人类反馈强化学习

Abstract: The emergence of generative artificial intelligence （AI） represented by ChatGPT signifies the comprehensive arrival of a new era of technology. The ChatGPT algorithm appears to possess a certain “power” that continuously “dictates” peoples lives. Simultaneously， heavy dependence on algorithmic technology is causing a continuous erosion of human rights， leading to the phenomenon of “alienation of the subjectobject relationship”. As a leading technology in natural language processing， the ChatGPT algorithm not only represents an advanced data analysis tool, but also exhibits a tendency towards the exercise of power， profoundly influencing and controlling human cognitive patterns， behavioral modes， and even potentially dominating the operation of nations and societies. The entire process of power exercise and operation of the ChatGPT algorithm remains in a “black box” state， exhibiting a tendency that goes beyond， deviates from， and operates independently human rights， thereby posing a threat to human subjectivity. To mitigate the legal risks associated with the exercise of power by the ChatGPT algorithm， it is urgent to return to the fundamental principles of law. Starting from the basic principles of safeguarding rights with power， constraining power with rights， and balancing power with power， it is necessary to explore and construct new regulatory concepts and paths， conduct a comprehensive examination and reflection on the power of the ChatGPT algorithm across multiple domains， levels and dimensions.

Key words: ChatGPT; algorithmic power; subjectivity; reinforcement learning from human feedback