智能发展的结构循环_一种跨实现的描述框架 The_Structural_Cycle_of_Intelligence_Development

Engineering notes

Key topics: autonomous driving, large language model, prediction. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

【中文摘要】 智能系统如何在长期运行中持续发展,是认知科学与人工智能共同关注的问题。现有研究多围绕特定实现或局部机制展开,缺乏能够统一描述不同智能系统发展过程的分析框架。本文提出一个跨实现的描述框架,将智能发展理解为一种结构循环:系统内部结构持续参与后续结构的形成,从而推动能力演化。在此基础上,本文进一步区分机制与约束两个层面:前者说明结构循环如何形成与维持,后者解释相同机制为何呈现不同的发展形态,并提出评价函数能否进入结构循环并被系统自主改写是区分定界发展与开放发展的关键维度。本文以人类与大型语言模型作为两类实现,对框架进行逐环节分析,并提出若干可经验检验的预测。本文旨在提供一种统一描述不同智能系统发展过程的理论框架,而非针对某一具体实现的发展模型。 ――――――――――― [English Abstract] How intelligent systems continue to develop over long-term operation is a question of shared concern to cognitive science and artificial intelligence. Existing research has largely centered on particular implementations or local mechanisms, and lacks an analytical framework capable of uniformly describing the developmental process across different intelligent systems. This paper proposes a cross-implementation descriptive framework that understands the development of intelligence as a structural cycle: internal structures already formed within a system continuously participate in the formation of subsequent structures, thereby driving the evolution of its capabilities. On this basis, the paper distinguishes two levels, mechanism and constraint: the former accounts for how the structural cycle forms and is sustained, while the latter explains why the same mechanism gives rise to different developmental forms. It further proposes that whether the evaluation function can enter the structural cycle and be autonomously rewritten by the system itself is the key dimension distinguishing bounded from open-ended development. Taking humans and large language models as two contrasting implementations, the paper analyzes the framework component by component and offers several empirically testable predictions. The aim is to provide a unified framework for describing the developmental process of different intelligent systems, rather than a developmental model of any particular implementation.

Links and sources

Comments