基于具身内生驱动力的人工智能自主意识原型构建研究

Abstract Current mainstream large language models and automated agents only have the capability of data fitting and task execution, without subjective self-awareness, endogenous behavioral motivation or real autonomous consciousness. The traditional paradigm of "basic template + large-scale data training + self-modification of codes" cannot break through the barrier between simulated intelligence and real consciousness. To tackle the fundamental problems such as lack of self-boundary, native driving force and subjective cognition of existing artificial intelligence, this paper proposes a new AI construction system: embodied endogenous driving force, self-boundary cognition, memory evolution and self-referential reflection. Abandoning the pure text training mode, this system takes the physical robot body as the cognitive carrier, builds a bionic homeostasis reward and punishment mechanism, and establishes the cognitive boundary between self and the external world through continuous physical interaction. A hierarchical permission self-evolution framework is adopted to realize autonomous learning and iterative optimization under stable basic logic. This paper elaborates the core theory, hardware configuration, software layered architecture, implementation procedure and safety constraint mechanism of the system. It provides innovative ideas and operable schemes for the engineering practice of artificial autonomous consciousness.