Artificial Intelligence Is a Solver, Not a Subject 人工智能的求解器本质与主体性不可实现

Objective: To examine the mechanism of artificial intelligence as a computing system in a valid operating state, to establish an externally observable boundary between humans and computing systems, and to determine whether technological iteration can make artificial intelligence a subject. Method: The argument is developed from basic theories of computation. The halting problem is used to clarify the necessity of external observation for system output and operating state. Automata theory is used to explain the relation between external input, program execution, and state transition. On this basis, the operating process of computing systems is summarized as “solving,” and the mechanism of “accepting input—running programs—presenting externally observable states” is defined as “forced closure.” The Church–Turing thesis is further used to extend this constraint from individual Turing machines to all computable systems, including multi-agent and multi-machine systems. Result: A computing system in a valid operating state cannot initiate solving outside the given solving structure. After accepting valid external information input, it must enter a rule-governed process and present an externally observable closed state. Without new external information input, it cannot actively break its existing externally observable stable state. Therefore, artificial intelligence remains a solver constrained by forced closure. In contrast, human beings can suspend the relation between input and action, and can actively maintain or break an externally observable stable state. Conclusion: The observable human-machine boundary lies in suspension. Current technological routes, including generative models, chain-of-thought reasoning, tool use, retrieval augmentation, reinforcement learning, multi-agent collaboration, and embodied intelligence, only optimize solving capacity within the structure of forced closure. Within the theoretical boundary of computable systems, technological iteration cannot generate subjectivity in artificial intelligence. This conclusion provides a computational-theoretical basis for the human-centered principle in artificial intelligence governance. 目的：研究合法运行状态下计算系统的机制本质，建立人机可观测边界，并判断技术迭代能否使人工智能成为主体。基于上述回答，为人工智能治理原则提供计算理论依据。 方法：从基础理论出发进行推导。以图灵停机问题明确输出的可观测性；以自动机理论论证接受输入与程序运行的关联性，揭示“接受输入—强制运行程序—强制输出可观测状态”的机制，定义计算系统为“强制闭合”的“求解器”。以丘奇—图灵论题将约束范围扩展所有计算系统，包括多机体系。 结果和结论：通过是否接受输入/是否打破既有稳定状态的行动差异，明确了人类与计算系统的可观测边界。经对当前主流技术路线在理论基础和应用层面的逐项判定，明确了技术迭代无法突破“强制闭合”，人工智能无法成为主体。“以人为本”的治理原则有了计算理论支持。